Cool and quiet Ryzen APU build 15L

alexs

Chassis Packer
Original poster
Nov 20, 2018
17
30
Hi everyone!

A lurker here until now, I've finally finished my SFF build.
Ryzen 2400g, no GPU, Sharkoon QB one, and a modified Be Quiet 240 watercooling loop.
This is likely going to be my workstation for a while.



I got this case for cheap, I liked it and it kinda defined the build. It's got mounting holes for a 240mm rad, so I decided to make it water cooled. A radiator that large meant I could go with low RPM fans and ultimately, a very quiet system.




PSU choice was tricky. I really wanted a 12V solution with a DC-DC converter, but power connector location on B450 motherboards meant inductors of the converter would touch the outer RAM stick, not something I'd want to see. The case is meant for ATX or SFX PSUs, but neither option was particularly appealing. What I got instead was Be Quiet TFX2.

BQ TFX2 is a single-rail 12V supply with built-in DC-DC downconverters. The primary stage has claimed efficiency of ~90%, similar to Meanwell RPS series. It's much larger than RPS-200 and larger than RPS-300, but also much less crowded.

It's overpowered for the purpose, this system should not draw much more than ~100W or so. But it was easily available, reasonably cheap, and reading through the datasheet for RPS-200, I realized I could probably run it without the noisy 80mm fan — as long as I keep the temps in the case under control.

The board also happened to fit the case very well (with a 3D-printed support bracket).



The wires were shortened to size, and I dropped wire count for low-voltage lines. A very conservative estimate for current capability of a 18AWG wire is 7A, and there were something like five of them for the +5V and no less than four on +3.3V. No way this system would draw anything close to 30A on the +5V line.



The thinner bundle got somewhat manageable, but still extremely unpleasant. I really need to figure out something with DC-DC boards.



For cooling, I got BQ Silent Loop 240, for two reasons: standard G1/4 threads, and the small integrated pump. And it's supposedly quiet. Standalone pumps I could get were too large and too powerful for this case.

Now whatever little knowledge I have about water cooling (and water heating for that matter) tells me AIO without a reservoir is a bad idea. It can be done, but it's tricky, and it wouldn't fit in this layout anyway. Air has to be kept out of the pump if I want it to run quiet. There are numerous reports about BQ pumps making noises, and all of them have AIR IN THE SYSTEM written all over.

So I got some G1/4 fittings and mounted a Barrow L50-60 reservoir right onto the rad.

Stock BQ Pure Wings (~1800RPM max) got replaces with the slowest Scythe fans (~900RPM max) to reduce the noise further.



I though I would re-use BQ's tubing and the original compression fittings, but the tubes turned out to be of unusual size and couldn't be cut. So I ended up using 8/12mm silicone rubber tubing on 10/13mm Barrow fittings.



This layout should leave enough free air in front of the fans to make sure they work efficiently. The rule of thumb I've heard is one fan diameter of free air, but with 30mm rad and 25mm fans and the boards it just wasn't going to happen. Well, at least I've got like half of that.

Fans located like that also make sure any air that gets heated inside the case gets pulled out of the case quickly, and never recirculates inside. Cold air gets drawn from the bottom, front, right and possibly top sides of the case, and exits through the rad and through the left panel.




The cooling system turned out to work really well.

Idle: 32C, 1500MHz, fans 500RPM, barely audible.
Full CPU load: 55C, 3750MHz, fans 850RPM, some fan hum

It doesn't really want to go above 55C, and goes back to mid-30s quickly once the load is removed. I did not test iGPU loads yet however. The iGPU here could easily produce about as much heat as the CPU cores.

PSU seems to be working well with this kind of almost-passive cooling. The large radiator stays around 40-45C at all loads, and the hottest element turn out to be the transformer at 55C or so.

Pump noise is almost non-existant at 1700..1800 RPM, but turns into audible whine at 2200..2300 which is the top speed for this pump. Side note: ASRock's board config tool sucks.

The assembled case is REALLY HEAVY for its size. I mean, it's not exactly unexpected, but still.
 

fergunat000

Chassis Packer
Nov 27, 2017
15
5
How do you feel the CPU Block/Pump works? do you think it would have enough pressure head to add more components to the loop (i.e. GPU block), and can you tell the sound of the pump vs the fans or is there no audible difference?
Love you build1
 

owliwar

Master of Cramming
Lazer3D
Apr 7, 2017
586
1,082
oh wow
great job but boy that open psu frame is a bit scary. you seem to know what you're doing though
very very interesting
 

alexs

Chassis Packer
Original poster
Nov 20, 2018
17
30
do you think it would have enough pressure head to add more components to the loop (i.e. GPU block), and can you tell the sound of the pump vs the fans or is there no audible difference?

I did not bother checking how much it can do, but I see no reason why it wouldn't, in a small case at least.
It doesn't look particularly wimpy. Tube routing will likely be more of a problem.

Sound-wise, I would say it's quieter than the fans right now. It should be doing between 1700 and 1800 RPM, and the fans around 500 RPM. I need to put my ear right next to the case to hear it. At higher RPM it's going to be a bit louder. I'd say it's like having a 2.5" HDD in the case.
 
  • Like
Reactions: fergunat000

CC Ricers

Shrink Ray Wielder
Bronze Supporter
Nov 1, 2015
2,233
2,556
That's an interesting way to fit a PSU but I do like the DIY nature of it.

What's the monitor stand that you're using?
 

alexs

Chassis Packer
Original poster
Nov 20, 2018
17
30
Well. RIP BQ TFX2 2018-2018.
Blew fuses in a rather spectacular way while under moderate load.
Investigation revealed a shorted chopper MOSFET and apparently nothing else.

The passive-cooling experiment has to be declared inconclusive for now. The exact reason the switch failed is not clear, but reading through the datasheet, I'm pretty sure now this thing was nowhere near overheating.

The box is now re-built with Meanwell PPS-200-12 and PicoPSU.
I wanted to try a 12V build anyway, and it was a perfect opportunity.



Cable routing is so much easier now. Meanwell PSU runs a bit hotter than BQ, but not much. I'm picking around 50C on the transformer. Should be expected I guess, it's a lot smaller and more tightly packed. Despite my initial doubts, PicoPSU fits well next to the memory modules, with some space to spare even. On the downside, MW draws about 7W without load, way more than I'd expect. Sadly, not something that can be solved easily. Gotta think on that.

I had some hopes MW would be totally quiet, but it's not. Coil whine is quieter than with BQ but still audible, and also constant. BQ's was intermittent.
 

Boil

SFF Guru
Nov 11, 2015
1,253
1,094
Why not just a standard HDPlex 160W AC-DC & HDPlex 160W DC-ATX (or a KMPKT Dynamo Mini)...?
 

alexs

Chassis Packer
Original poster
Nov 20, 2018
17
30
The only reason to even consider HDPlex over 12V MW is the poor no-load performance of MW PPS, something I did not know before I assembled the system. Pretty much everything else in not in favor of HDPlex. Once I know that, there's EPP-200 which is 12V, about the same size as HDPlex 150W, claims the same <0.5W no-load power, and is much easier to get.

HDPlex system uses an extra 19V intermediate (AC -> 19V -> 12V) which I don't understand. It's just not a good idea.